memory requirement

[zz8615]

i am trying to train the model using 2 Telsa V100. I have changed the batch_size to 1 but it still tells me it is out of memory. Is this expected? Thanks!

[XavierCHEN34]

You coul use "ddp_sharded", which requires smaller memories
trainer = pl.Trainer(gpus=1, strategy="ddp_sharded", precision=16, accelerator="gpu", callbacks=[logger], progress_bar_refresh_rate=1)

[mo230761]

Hello，I would like to ask if the code can run using this method. Thank you!